pgloom 0.3.0

Postgres-backed reusable workflow and task orchestration runtime.

pgloom

pgloom is a Postgres-backed workflow and task orchestration runtime for building domain-specific automation systems. It provides durable workflow state, task dependencies, worker leases, slots, approvals, idempotency records, model usage accounting, artifacts, notifications, health checks, and a scenario harness.

Quickstart

pip install pgloom
createdb pgloom_dev
export PGLOOM_DATABASE_URL=postgresql://localhost/pgloom_dev
pgloom db migrate
pgloom workflow create --domain demo --name smoke
pgloom task enqueue --workflow-id <workflow_id> --slot fake --task-type fake.complete
pgloom worker run-once --slot fake

For local development:

scripts/bootstrap_dev_env.sh
source .venv/bin/activate
just test

Concepts

flowchart LR
  W[Workflow] --> T[Tasks]
  T --> D[Dependencies]
  T --> S[Slots]
  S --> R[Workers]
  T --> H[Handlers]
  H --> E[Events]
  H --> A[Artifacts]
  H --> M[Model usage]

• Workflows group related tasks and carry domain metadata.
• Tasks are durable units of work with state, attempts, payloads, results, blockers, dependencies, and leases.
• Slots model constrained execution capacity such as claude, codex, or qa.
• Handlers are application code that claims tasks and returns structured outcomes.

Embedding

Domain orchestrators import pgloom and keep their own domain behavior outside the core runtime. engineering-orchestrator is the reference consumer planned for engineering-specific planner, implementer, reviewer, QA, historian, GitHub, BRAID, worktree, and Telegram behavior.

Documentation

• Architecture
• Postgres schema
• Scenario harness
• Release process